Skip to content
Agent Process

Connect the agents you use

Connect your AI agent to Agent Process once. Your processes, data, emails and approvals are then ready in every conversation, with nothing new to learn.

19 agents supported, each with a step-by-step setup.

Claude Desktop

Connected to Agent Process

What processes can you see?

Ask Claude Desktop anything…
Illustrative conversation. Same setup in every agent.

Connect in a few minutes

A few minutes, once. During the pilot we do it with you on a call.
  1. Create an access key

    In Agent Process, open Agents and create a key. You choose whether it can only run processes, or build them too.

    Key for my agent

    ••••••••••••••••

    Run processes

  2. Paste one block of settings

    The Agents page fills in your address and key for you. Copy it into your agent’s settings.

    "mcp-chats": {
      "url": "your address",
      "token": "your key"
    }

    Simplified. Each agent’s setup page shows its exact settings.

  3. Ask your agent

    It lists the processes it can run for you, and you’re ready to hand over work.

    What processes can you see?

    Refund request, supplier onboarding and 1 more.

Find your agent.

Every setup is written from that agent’s own documentation.

Personal agents

AI assistants that work on your behalf.

Coding agents

Agents in the terminal, for technical teams.

Editors and desktop apps

Desktop apps and editors your team already has open.

Your own code

Build Agent Process into your own software.

Later: Claude and ChatGPT, by signing in

Add Agent Process in Claude or ChatGPT, sign in and choose what the agent may do. No key to copy. Not available in the pilot yet; until then, connect with an access token.

It works within your limits.

A connected agent acts as you, with the access you chose. You can revoke its key at any time.

It can

  • See the processes published for you
  • Start runs and do its steps
  • Use the data, systems and email templates you granted
  • Ask the right person to approve

It can’t

  • See your passwords or API keys
  • Change data through an action you haven’t enabled
  • Send emails you didn’t write, or to people you didn’t allow
  • Change Jev’s answers or the run’s record

These limits cover work routed through Agent Process. Tools your agent has elsewhere are outside our control.

Connecting in the app

The Agents page walks through it, and every run an agent starts is listed.
Connecting Claude Code: choose the agent, create a token, then run one command
Connecting Claude Code: choose it, create a token, run one command.
Runs of the refund process, newest first, active or waiting
Every run of a published process, newest first.
An agent's token with its access, when it was last used, and a Revoke button
See when each token was last used, and revoke it at once.

Then just ask.

Plain requests, in the chat you already use.
  • Design a process

    “Design a process for onboarding a new supplier: collect their details, check them against our vendor list, and get finance to approve.”

  • Start a run

    “Start supplier onboarding for Brightwell Packaging Ltd.”

  • Do agent steps

    “Pick up any agent steps that are ready and do them.”

  • Check on runs

    “Which runs are waiting on someone, and for what?”